부산대학교 도서관

Cardiovascular diseases (CVDs) continue to be a major concern in the medical field to date. Among the many diagnostic tools, electrocardiogram (ECG) remains one of the main ways with which to detect cardiac abnormalities. Although classic ECG interpretation is powerful, it is a laborious and time-demanding task and tends to be onerous for physicians. This study suggests a means of effective machine learning to increase the accuracy and speed of diagnosing CVDs based on ECG signals and other sensor data. On the other hand, we are concerned with the application of Frequency Domain Measures in deep-learning methodologies of image classifiers. Other models include the raw ECG sent to a 1-dimensional convolutional model together with the XGBoost model for the time-series feature-based prediction. Preliminary assessment reveals that Frequency Domain Measures provide good atrial fibrillation prediction but are quite limited for other types of arrhythmias. Computational results for the XGBoost model are, however, efficient for predictions on long data but proceed at the cost of extended inference times, considering the computational demand of the pre-processing stage. This paper opens up the scope of the potential for machine learning in revolutionizing ECG-based CVD detection, and we have to keep on the protocols of refinement and research toward this usage.